Smarter Infrastructure Planning - Overcoming the 3-Month Billing Data Gap

Written By Gareth Williams

Accurate peak-day demand profiles are essential to the NPV calculations underpinning multi-million-dollar water network investments.

The peak day hydraulic modeling scenario determines available capacity and whether new infrastructure is required to support proposed development.

Infrastructure planners therefore need to understand how customers use water on a single day when the network is stressed.

Customer billing data, however, only shows total consumption over three-month periods.

To fill this gap, engineers turn to SCADA subsystem data to build diurnal profiles - but these aggregate all demand categories (houses, apartments, commercial, etc.) into a single profile.

We’ve been working with an infrastructure planning group that wanted to do a regression analysis on their big data to overcome this limitation.

The analysis outputs evidence-based, 30-minute profiles for different demand categories, Houses and Apartments on a peak day, for example.

With this clearer view of residential demand behavior, the planning team gained deeper insight into how peak flows occur across their system.

Smart Meters Are Not (Yet) the Solution

Australian utilities and councils are investing heavily in smart meters - a good move, given the evident impact on water conservation when customers are better informed about their usage.

However, in the context of peak demand profiling, smart meter coverage remains below 10%, and the available datasets are still too short to capture peak-day events.

That means smart meters can’t yet reveal the peaks that drive infrastructure sizing, although they remain a valuable line of evidence for decision-making.  

Fortunately, utilities possess a rich source of historical SCADA data, that when leveraged by a big data tool, offers a deeper understanding of peak behavior right now.

How the Regression Works

The regression approach allows for building a diurnal demand profile - for Houses and Apartments - for any day in the historical SCADA records e.g. Peak Day.

For each subsystem, the model develops relationships between measured subsystem flows and the counts of customer categories to solve for the average and peak contributions from different customer types. The equations below are applied at 30-minute timesteps to output diurnal profiles.

Average Day Equation:

𝑄𝐴𝑣𝐻 x π»π‘œπ‘’π‘ π‘’π‘π‘œπ‘’π‘›π‘‘ + 𝑄𝐴𝑣𝐴𝑝 x π΄π‘π‘Žπ‘Ÿπ‘‘.π‘π‘œπ‘’π‘›π‘‘ + π‘„π‘π‘œπ‘š + 𝑄𝐼𝑛𝑑 + π‘„π‘œπ‘‘β„Ž. + π‘’π‘›π‘Žπ‘. = 𝑆𝐢𝐴𝐷𝐴_π΄π‘£π‘„π‘ π‘’π‘π‘ π‘¦π‘ π‘‘π‘’π‘š (1)

Peak Day Equation:

π‘„π‘ƒπ‘’π‘Žπ‘˜π» x π»π‘œπ‘’π‘ π‘’π‘π‘œπ‘’π‘›π‘‘ + π‘„π‘ƒπ‘’π‘Žπ‘˜π΄π‘ x π΄π‘π‘Žπ‘Ÿπ‘‘.π‘π‘œπ‘’π‘›π‘‘ + π‘„π‘π‘œπ‘š + 𝑄𝐼𝑛𝑑 + π‘„π‘œπ‘‘β„Ž. + π‘’π‘›π‘Žπ‘. = 𝑆𝐢𝐴𝐷𝐴_π‘ƒπ‘’π‘Žπ‘˜π‘„π‘ π‘’π‘π‘ π‘¦π‘ π‘‘π‘’π‘š (2)

Where:

  • 𝑄𝐴𝑣𝐻 and 𝑄𝐴𝑣𝐴𝑝 are the unknown average flow contributions per House and per Apartment

  • π‘„π‘ƒπ‘’π‘Žπ‘˜π» and π‘„π‘ƒπ‘’π‘Žπ‘˜π΄π‘ are the unknown peak flow contributions per House and per Apartment

  • Qoth. represents β€˜other’ demand categories, and

  • β€œunac.” represents unaccounted-for water, also known as non-revenue water.

SensorClean’s regression module provides flexibility in selecting the date ranges used to define subsystem averages, allowing for trend in the data commonly driven by development.

If connection numbers are available over time, trends can be removed to understand changing water behaviors - independent of growth - in the subsystem.

Using regression in this way has been discussed in the water industry before, but widespread adoption has been limited by the practical constraints of business-as-usual spreadsheets - such as the one-million-row limit - and the realities of large SCADA data volumes.

Diversity Factors and Peaking in Water Distribution Systems

In hydraulic modelling, each subsystem’s peak-day demand is usually represented by a peaking factor - the ratio of maximum to average water use.

However, when many subsystems are combined, their peaks don’t necessarily occur at the same time. The sum of individual peaks exceeds the actual system peak.

Smaller subsystems tend to experience greater variation in demand because there are fewer customers to smooth out individual behavior. This is captured by a diversity factor depending on the number of connections.

As systems grow larger, the timing of peaks becomes more spread out - one area’s morning peak might offset another’s afternoon lull.

Larger systems therefore exhibit fewer coincident peaks and smaller overall peaking factors, while smaller systems can show sharper peaks.

Looking Ahead

By leveraging existing SCADA data and regression techniques, utilities can build evidence-based peak demand profiles today - without waiting for universal smart meter coverage.

Get in touch to see how SensorClean can bring your SCADA data to life and support data-driven infrastructure planning.

Gareth Williams
Next

Subsystem Water Balances That Drive Smarter Investments